Google releases Imagen 2 model: Achieving high-definition AI image generation and multilingual rendering

Google has unveiled the mystery behind its text-to-image artificial intelligence model, Imagen 2. This is the latest version of its text-to-image AI model. This new model is expected to achieve more realistic and detailed image generation capabilities through advanced neural network technology.

Unlike OpenAI's DALL-E3 or Adobe and Midjourney's tools, Google focuses on providing APIs for developers to use, rather than standalone consumer applications.

Building on the existing Imagen API, Imagen 2 has made significant improvements in image quality and text prompt understanding. Through improvements in training data and methodology, the images generated by Imagen 2 have higher resolution and more visually appealing details that better match the provided descriptions.

Specifically, Google has enhanced the image captions used to train Imagen 2, helping the model better grasp context and subtle differences. Additional training sets focus on improving Imagen 2's rendering of challenging areas such as hands and faces, while reducing visual artifacts. The company has also implemented an image quality scoring system to further optimize the output.

Imagen 2 introduces other new features that allow for better control of image attributes. Users can now provide style reference images, and Imagen 2 is able to adopt requested styles such as lighting, texture, and color palette.

Imagen 2 can directly generate new content onto the original image through inpainting and outpainting capabilities.

The API has also added advanced inpainting and outpainting capabilities, which allow generated content to be inserted into existing images or extend the image beyond its boundaries.

Multi-language support allows for prompts and outputs in seven languages so far, with more to come in the future. Imagen 2 can even render text within the image in the appropriate language.

This provides rich possibilities for brand creation and localization. Logo generation allows users to create custom logos that can be seamlessly integrated into other media.

Google emphasizes responsibility as the core of Imagen 2 development. Prior to release, the company conducted security testing for sensitive categories to avoid issues. Imagen 2 is also connected to Google's SynthID tool, which applies imperceptible watermarking to AI-generated images at the pixel level for authentication and tracking.

Imagen 2 is now available through Google's Vertex AI platform for whitelisted paying customers. Since its launch, major creative brands including Snap, Shutterstock, and Canva have become early adopters.